4片bytearray使用多線程進行比較

Question

我被困在項目的第二階段：將一個byte []拆分為4個切片（加載QuadCore I5 CPU），然后在每個切片上，在每個核心上啟動一個線程（比較任務）。

原因是試圖加快兩個相同大小的bytearrays之間的比較我如何線程化？

    [DllImport("msvcrt.dll", CallingConvention = CallingConvention.Cdecl)]
    static extern int memcmp(byte[] b1, byte[] b2, long count);




class ArrayView<T> : IEnumerable<T> 
    {
        private readonly T[] array;
        private readonly int offset, count;
        public ArrayView(T[] array, int offset, int count)
        {
            this.array = array; this.offset = offset; this.count = count;
        }
        public int Length { get { return count; } }
        public T this[int index] {
            get { if (index < 0 || index >= this.count)
                throw new IndexOutOfRangeException();
            else
                return this.array[offset + index]; }
            set { if (index < 0 || index >= this.count)
                throw new IndexOutOfRangeException();
            else
                this.array[offset + index] = value; }
        }
        public IEnumerator<T> GetEnumerator()
        {
            for (int i = offset; i < offset + count; i++)
                yield return array[i];
        } 
        IEnumerator IEnumerable.GetEnumerator()
        {
            IEnumerator<T> enumerator = this.GetEnumerator();
            while (enumerator.MoveNext())
            {
                yield return enumerator.Current;
            }
        }
    } 




    public void CopmarArrSlice()
    {

    byte[] LoadedArr = File.ReadAllBytes("testFileCompare2Scr.bmp");

    int LoddArLn = OrgArr.Length;
    int range =  (LoddArLn / 4) - LoddAremainder;
    int divisionremain = LoddArLn - (range * 4);

    ArrayView<byte> LddP1 = new ArrayView<byte>(OrgArr, 0, range);
    ArrayView<byte> LddP2 = new ArrayView<byte>(OrgArr, p1.Length, range);
    ArrayView<byte> LddP3 = new ArrayView<byte>(OrgArr, (p1.Length + p2.Length), range);
    ArrayView<byte> LddP4 = new ArrayView<byte>(OrgArr, (p1.Length + p2.Length + p3.Length), range + divisionremain);


        if (AreEqual(LddP1, CapturedP1)) ....Do Somthing

    }


    public bool AreEqual(byte[] a, byte[] b)
    {
        if (a == b)
           return true;
        if (a == null || b == null)
            return false;
        if (a.Length != b.Length)
            return false;
        return memcmp(a, b, a.Length) == 0;
    }


    CopmarArrSlice();

在這種情況下，如何使用AreEqual（使用memcmp）將其與使用4個線程/ Parallelism進行比較，以計算每個CpuCore

Answer 1

我編寫了一個在可能的情況下利用多個核心的函數，但它似乎遭受了p / invoke調用的嚴重性能損失。 我認為這個版本只有在測試非常大的數組時才有意義。

static unsafe class NativeParallel
{
    [DllImport("msvcrt.dll", CallingConvention = CallingConvention.Cdecl)]
    static extern int memcmp(byte* b1, byte* b2, int count);

    public static bool AreEqual(byte[] a, byte[] b)
    {
        // The obvious optimizations
        if (a == b)
            return true;
        if (a == null || b == null)
            return false;
        if (a.Length != b.Length)
            return false;

        int quarter = a.Length / 4;
        int r0 = 0, r1 = 0, r2 = 0, r3 = 0;
        Parallel.Invoke(
            () =>  {
                fixed (byte* ap = &a[0])
                fixed (byte* bp = &b[0])
                    r0 = memcmp(ap, bp, quarter);                        
            },
            () => {
                fixed (byte* ap = &a[quarter])
                fixed (byte* bp = &b[quarter])
                    r1 = memcmp(ap, bp, quarter);
            },
            () => {
                fixed (byte* ap = &a[quarter * 2])
                fixed (byte* bp = &b[quarter * 2])
                    r2 = memcmp(ap, bp, quarter);
            },
            () => {
                fixed (byte* ap = &a[quarter * 3])
                fixed (byte* bp = &b[quarter * 3])
                    r3 = memcmp(ap, bp, a.Length - (quarter * 3));
            }
        );
        return r0 + r1 + r2 + r3 == 0;
    }
}

在大多數情況下，它實際上比優化的安全版本慢。

static class SafeParallel
{
    public static bool AreEqual(byte[] a, byte[] b)
    {
        if (a == b)
            return true;
        if (a == null || b == null)
            return false;
        if (a.Length != b.Length)
            return false;

        bool b1 = false;
        bool b2 = false;
        bool b3 = false;
        bool b4 = false;
        int quarter = a.Length / 4;
        Parallel.Invoke(
            () => b1 = AreEqual(a, b, 0, quarter),
            () => b2 = AreEqual(a, b, quarter, quarter),
            () => b3 = AreEqual(a, b, quarter * 2, quarter),
            () => b4 = AreEqual(a, b, quarter * 3, a.Length)
        );
        return b1 && b2 && b3 && b4;
    }

    static bool AreEqual(byte[] a, byte[] b, int start, int length)
    {
        var len = length / 8;
        if (len > 0)
        {
            for (int i = start; i < len; i += 8)
            {
                if (BitConverter.ToInt64(a, i) != BitConverter.ToInt64(b, i))
                    return false;
            }
        }
        var remainder = length % 8;
        if (remainder > 0)
        {
            for (int i = length - remainder; i < length; i++)
            {
                if (a[i] != b[i])
                    return false;
            }
        }
        return true;
    }
}

Answer 2

我認為你不必在單個線程和經典的c＃方式中分割你的字節

foreach(byte currentArr in LoadedArr)
{
    if (AreEqyal(currentArr, CapturedP1))
           ....Do Somthing
}

但是要通過將工作負載分配給多個線程來處理每個字節，您必須使用以下語法;

// max your threads count in my case 16,
int[] sums = new int[16];// optional, just to know the workload
public void ProcessMyByte(byte current)
{
    if (AreEqyal(current, CapturedP1))
           ....Do Somthing

       // optional just to know what thread is in
        sums[Thread.CurrentThread.ManagedThreadId]++;// increment the number of iterations done by the thread who did this elementary process
}

.... Main()...
{
....

byte[] LoadedArr = File.ReadAllBytes("testFileCompare2Scr.bmp");
Parallel.ForEach(LoadedArr, ProcessMyByte);

...
}

所以並行性將代表你管理，甚至更好，因為當一個線程處於空閑狀態時，它會獲得下一個任務，而不是像你將它分成4個，每個線程必須處理Length / 4。